Methods and Apparatus for Detecting, Filtering, and Identifying Objects in Streaming Video

2. The method of claim 1, wherein generating a first triplet in the plurality of triplets comprises classifying an image of the first object as the reference data point of the first triplet and classifying an image of the second object as the positive data point of the first triplet.

3. The method of claim 2, wherein the first object is visually similar to the second object.

4. The method of claim 2, wherein the first object is visually dissimilar to the second object.

5. The method of claim 2, wherein the first object has a non-stationary similarity distribution with respect to the second object.

6. The method of claim 2, wherein the first object has an extrinsic similarity with respect to the second object.

7. The method of claim 6, wherein the extrinsic similarity of the first object with respect to the second object is based on at least one of current trends, user interactions, related purchases, or geographical location.

8. The method of claim 6, wherein the first object is a first piece of clothing and the second object is a second piece of clothing typically worn with the first piece of clothing.

11. The method of claim 1, wherein generating a first triplet in the plurality of triplets comprises classifying an image of the first object as a reference data point of the first triplet and an image of the third object as a negative data point of the first triplet.

12. The method of claim 11, wherein the first object is visually similar to the third object.

13. The method of claim 11, wherein the first object is visually dissimilar to the third object.

14. The method of claim 11, wherein the first object has a non-stationary similarity distribution with respect to the third object.

15. The method of claim 11, wherein the first object has an extrinsic similarity with respect to the third object.

16. The method of claim 1, wherein generating the plurality of triplets comprises generating between about 5×106 triplets to about 1.25×1012 triplets.

17. The method of claim 1, wherein generating the plurality of triplets comprises generating the plurality of triplets from a plurality of images containing between about 2000 total images and about 1,000,000 total images and at least 4 images per class of similar object.

18. The method of claim 1, wherein calculating the loss comprises calculating at least one of an N-pair loss, a triplet loss with L1 norm, a triplet loss with L2 norm, a lifted structure loss, or a margin-based loss.

20. The method of claim 1, wherein returning the representation of the object similar to the object appearing in the image shown in the display comprises retrieving the representation from a database.

21. The method of claim 1, wherein returning the positive data point, the reference data point, and/or the negative data point are taken from at least one of an image, a waveform representation of an audio clip, or a bag-of-words representation of text.

23. The system of claim 22, wherein the first object is visually similar to the second object.

24. The system of claim 22, wherein the first object is visually dissimilar to the second object.

25. The system of claim 22, wherein the first object has a non-stationary similarity distribution with respect to the second object.

26. The system of claim 22, wherein the first object has an extrinsic similarity with respect to the second object.

27. The system of claim 22, wherein the plurality of triplets comprises between about 5×106 triplets to about 1.25×1012 triplets.

28. The system of claim 22, wherein the plurality of triplets is generated from a plurality of images comprising between about 2000 total images and about 1,000,000 total images and at least 4 images per class of similar object.

30. The system of claim 29, wherein the representation is an image, the server is configured to return the image in response to the query from the user and the user interface is configured to display the image to the user.